This analysis presents the most comprehensive set of accelerometer-derived indicators of physical activity/inactivity using the largest nationally representative data set yet available. As a nation (excepting pregnant women), U.S adults average 6,564 ± SE 107 censored steps/day, and after considering non-wear time, they spend approximately 56.8% of the rest of the waking day in sedentary time, 23.7% in low intensity, 16.7% in light intensity, 2.6% in moderate intensity, and 0.2% in vigorous intensity. Overall, approximately 3.2% of U.S. adults achieve public health guidelines.
Examined across BMI categories for both sexes, a consistent decreasing and statistically significant gradient is apparent for all physical activity volume indicators, as is expected. With regards to time indicators (i.e., measured time spent in various intensity categories), there was a clear moderating effect of sex. Significant gradients were observed for non-wear time, sedentary time, light time, moderate time, and vigorous time (but not low intensity time) in males. For females, only gradients in moderate and vigorous intensity time were significant. Further, despite the fact that achievement of public health guidelines, is, overall, a rare phenomenon (at least as measured by the ActiGraph accelerometer), a decreasing gradient was still evident across BMI categories. Strath et al.  have also reported a relationship between time in MVPA and markers of obesity in U.S. adults using the 2003-2004 NHANES data. As this is a cross-sectional analysis, we are not able to conclude whether these decreased physical activity indicators lead to overweight and obesity or whether weight gain reduces these physical activity indicators. Other evidence suggests that both mechanisms are at work on an individual level [29, 30]. Body weight status is a complex function of other contributors, including dietary intake, although this is not a focus of this analysis.
The observed gradient in breaks in sedentary time was minimal (a difference of 1-2 breaks/day across BMI categories) and only statistically significant in males. Therefore, although these are only cross-sectional data, these minimal differences would suggest that this specific aspect of sedentary behavior may not be an important contributor to the obesity epidemic  (but this does not negate a possible contribution to other important health-related outcomes). Previous reports linking sedentary behavior to obesity have used questionnaire methods to recall sitting time [32, 33], but also the same brand and model of accelerometer as used by NHANES [21, 34]. We also used the same definitions of sedentary time  and breaks in sedentary time  as these previous accelerometer-based reports so the very minimal differences we observed cannot be explained by differences in either instrumentation or cut point choices.
We observed a similar decreasing gradient across BMI categories in rate indicators, however, these are ratios and as such need to be interpreted cautiously since they are affected simultaneously by both the numerator and the denominator (in this case, time monitored by accelerometer). Although we observed no consistent differences in non-wear time across BMI categories, variation in either the physical activity volume indicator or time that the accelerometer was worn, or both, can distort conclusions. That being said, a post-hoc analysis of covariance was performed to adjust mean time spent in each intensity for wear time. The results of this analysis were very similar to the unadjusted results. For example, when examining time spent in the sedentary intensity for males, the trend across BMI categories stayed significant with a p-value of 0.0001 compared to the unadjusted p-value of 0.007. Furthermore, the analysis for time spent in the low intensity produced a p-value of 0.64 when adjusted for wear time and a p-value of 0.78 when not adjusted. All categories were examined and were found to produce the same conclusions as the unadjusted model.
Activity counts corresponding to a MET-defined moderate intensity physical activity appear to be much lower in older and overweight/obese populations. Specifically, Lopes et al.  conducted an ActiGraph calibration study and determined that 1,240 activity counts/minute represented the threshold for moderate intensity activity in such a population, a value that is less than what has been conventionally used to describe the same intensity behavior in this and other NHANES analyses (i.e., 2020 activity counts/minute) . Physical activity is a behavior and it is quantified herein objectively as steps taken or time above a specific activity count threshold; this threshold captures movement as acceleration. Although conclusions about the metabolic cost of this behavior appear to be affected by factors known to influence energy expenditure (e.g., body mass), we remain nonetheless confident that differences (or lack of differences) between BMI-defined weight categories in objectively-monitored steps taken or their acceleration are real. To emphasize, energy expenditure is higher in obese individuals due to their higher body mass . In terms of physical activity, however, we found that obese individuals (regardless of sex) take fewer steps/day and spend less time in moderate and vigorous intensity activity.
It was not surprising that overweight and obese individuals tended to take fewer steps/day and that normal weight people tended to take more steps/day. As is evident from Figures 3 and 4, within each step-defined activity group, the percent of normal weight, overweight, and obese must add up to 100%; along the physical activity continuum the normal/overweigh/obese gradient "switches" from an upwardly sloping gradient to a downwardly sloping gradient. However, we did not expect to find that more overweight individuals were classified as somewhat active (i.e., taking 7,500-10,000 steps/day) compared to both normal weight and obese individuals. It is plausible that overweight individuals were more likely to modify their physical activity to affect their weight, and then this collective behavior was picked up as a distortion to the expected gradient. A previous analysis of 1-year tracking of pedometer-determined physical activity showed that the percent of obese individuals who increased their physical activity over the previous year was higher than those who decreased their behavior; further, as a group the obese were less stable (that is, more change occurred) in their behavior compared to a normal weight group . The relative instability of physical activity behavior by BMI category requires more research for confirmation.
In keeping with surveillance of reported leisure time physical activity  and inactivity  that show sex-specific differences, NHANES males were consistently more physically active (i.e., physical activity volume and rate indicators were higher and they spent more time in light and moderate intensity activity) and less physically inactive (i.e., they spent less time in non-wear time and low intensity time) than females across BMI categories. A consistent pattern in sedentary time was not evident between sexes across BMI categories. However, across these same categories, females took relatively more breaks in sedentary time compared to males. There were also relatively more males classified in the somewhat active, active, and highly active step-defined categories whereas relatively more females were classified in the basal activity, limited activity and low active categories. No other patterns (i.e., consistent results between sexes across BMI categories) were apparent.
Acknowledging the limitations of cross-sectional analysis, by scrutinizing a full panel of concurrent estimates of physical activity/inactivity across BMI-defined weight categories, we can begin to identify specific activity parameters that maximally differentiate between normal/overweight/obese samples and therefore best inform on-going surveillance efforts and physical activity interventions. An important caveat to keep in mind, however, is that this was a population analysis and the results do not necessarily apply to all individuals. Causality can only be substantiated, however, in longitudinal and intervention study designs.